A small story about Carrefour, PayPal, and algorithmic customer support.
Carrefour case
Given
I got my discount card in a local store in Tbilisi. It was linked to my phone number.
I kept using it for more than 4 years and never registered or used the Carrefour mobile application.
WHEN I decided to check how many loyalty points I had on my card
THEN I downloaded the Carrefour application from Google Play.
*Note: the login form asks for your email, but nobody asked for it in the store when the card was issued. So I chose to sign up.
Improvement idea:
Option to log in with a phone number + OTP.
WHEN I added my phone number in the sign-up form
THEN the validation message notified me that this number was already registered in the system
AND the sign-up flow was not blocked
AND the account was still created
AND the loyalty card number assigned in the application was different from my existing one.
*Note: another attempt to register a different email with the same phone number was stopped by validation in the sign-up form. Validation works for normal cases.
Why I’m upset? Well, there is no path to update or link the exiting card at that point. And for some reason, despite the fact that the system recognized my phone number, the mobile application assigned another card to it.
Support
Support in the mobile app is not actionable.
Support in the web version leads first to a knowledge base and then suggests calling a non-local phone number.
What probably happened
My guess is simple. Carrefour likely had a store-based loyalty system for years. Then they built a mobile application on top of it. And an identity mismatches appeared, and the automated support architecture simply isn’t designed for these edge cases, because this case in general quite “cheap” for a customer – “Okay, just use the card from the app” .
Customer support
Early internet services had email-based support. You wrote a message, and someone answered in a day or two.
That model collapsed once scale exploded. Emails pile up indefinitely, require long explanations, and create unpredictable workloads.
The next stage was call centers. Many banks and airlines still use them. But calls are expensive. A single phone interaction can cost several dollars. Multiply that by millions of users and the budget becomes enormous.
So the modern approach evolved into something that looks a bit cold from the outside: support deflection. The system tries to solve the issue before a human ever appears.
It usually looks like this:
First layer: knowledge base articles. The assumption is that most problems are common (for instance, as Carrefour and PayPal do)
Second layer: automated flows. You upload a document, the system checks it and approves or rejects it automatically (again PayPal).
Third layer: decision trees or bots.
Fourth layer — sometimes: human support, usually hidden behind several steps.
PayPal is a good example of this design philosophy. I recently had a case when I needed to confirm my identity. The system gives you very few options: either upload documents that match their requirements, or try calling support abroad.
Large companies like PayPal increasingly use what is called a “self-service first architecture.” They deliberately design support systems to reduce the probability of contact. Internally this is called the contact deflection rate. A successful system may resolve 80–95% of issues without human intervention.
From a business perspective this makes perfect sense.
From a user perspective, of course, it often feels like hitting a wall.
A bit of philosophy
Byung-Chul Han writes that modern society celebrates individuality and freedom, and the result is often radical isolation. In The Burnout Society and The Expulsion of the Other, he argues that modern systems remove friction, remove negativity, remove conflict — and in doing so they also remove the Other.
The “Other” in philosophy simply means another human presence who can resist, respond, or negotiate.
In older institutions the Other was unavoidable. A bank clerk could disagree with you. And a customer support agent could bend the rules. There was friction, yes — but also dialogue.
Algorithmic institutions work differently: they do not argue, they do not negotiate and they do not interpret context. They simply check conditions.
WHEN document valid,
THEN proceed
WHEN document invalid,
THEN reject.
From the system’s perspective this is elegant and efficient. But you are not speaking to anyone and you are simply trying to satisfy a rule set.
Han might say that modern systems eliminate the “Other” and replace it with transparent processes. And transparency sounds positive — but it also removes ambiguity, interpretation, and human mediation.
My PayPal and Carrefour stories are small examples of this broader transformation. We moved from personal authority to bureaucratic rules and then to algorithmic governance.
Customer support used to be a human conversation. Now it is often an algorithmic ritual where the user proves their existence to a machine.
Individuals wanted freedom, a freedom from institutions, hierarchy and authority. And now many of our interactions with institutions happen in complete solitude: no clerk, no negotiation, no shared space. Just a lonely individual facing a cold system.
Sometimes it works beautifully (especially if you consider yourself introvert 🙂 . Sometimes it doesn’t, like in my Carrefour case.
Back to Carrefour
My Carrefour card story is essentially a small identity paradox.
I exist twice:
- once as the loyalty card customer from four years ago
- once as a new digital account in the application
Both identities are “me”, but the system doesn’t know how to reconcile them. And honestly, I have no idea how to fix it. At this point it is simply easier to start using the new card generated in the application.
Maybe this is simply the price we pay for systems that must serve millions of users.
Leave a Reply